Artificial Intelligence at XXXL Group | Part 1
Author: Klaus Puchner, Program Manager AI & Team Lead
Currently an AI wave hits every industry with wishful promises, outperforming other methods of decision making and data visualization. Traditional companies face a new technology stack that is only little compliant to existing enterprise architecture standards. The importance of data turned everything designed to be stateless into a consuming data hub.
Luckily we already have usable data for machine learning systems and therefore it’s much easier for us to create data pipelines that allow for experiments and for a valuable contribution to their surrounding systems. We have always been a data driven company. Providing services to customers all around Europe, XXXLutz shapes the space most important to us – our homes.
As for any strategy decision, how we proceed in the field of AI and how we prepare ourselves for the opportunities and challenges that come with it is of the utmost importance. The major internet players provide different tools and platforms for processing data and for diving right into the technology – but this simply is not enough on our path to transforming into a technology company: AI needs to be at the heart of what we do.
IXI – our information extraction intelligence platform
When we started our AI initiative last May, it became obvious pretty fast that we needed an AI tech stack that can help shape the mindset of our company. We want to spark interest among our stakeholders. Also, the opportunities must be conveyed to those who profit most of it – the people who run our business. So we created a platform based on open source tooling that is capable of addressing different challenges:
We need to be able to experiment in a save space but also to adhere to the high quality standards we live by
AI services must be integrated by other services into our architecture
The handling of models and data needs to scale at the same pace as we do as a company
We care about usage workloads and our systems’ resource footprint
Similar use cases appear within the organisation and we want to add value to all of them
We still like to work with other platforms and use the best tools for the job at hand
Additionally, we must convey our standards on quality and simple usability to the people integrating our services, especially colleagues who develop and use the platform services.
To create a foundation that serves all those requirements, we decided to get best practices from open source software as building blocks for creating a focused AI software stack running on our infrastructure. To brand and market the platform, we gave it a name: “Information Xtraction Intelligence” (short form IXI).
The real MVP: IXI v1.0
The first version of IXI provided the solid foundation for future development and improvement of the platform. For the initial version we wanted to
integrate third party AI services
have a company-wide platform and playground to showcase available model functionality
provide implemented external AI model functionality via API (object recognition)
provide product owners with management information in a dashboard with statistics
reuse the existing UI patterns and component libraries of our online shop frontend
have a save user management and role-based authentication functionality
The frontend technology stack accelerated the concept phase and development immensely as we could use existing ReactJS components and business logic like routing. To scale with high volume, the document-based MongoDB allowed for structured storage even when under pressure by high parallel-processing in aiohttp (Python). For the initial version, we integrated two third party AI services: Moonvision and Google cloud vision and were able to get our first trained model with object detection for the first use cases and opportunities for experiments.
Whats next: IXI v2.0
IXI will be furnished with multiple new features in the releases to come. We are an international company that interacts with many countries, regions and languages. Accordingly, our technology stack for building AI models will be bilingual: Python (Jupyter) and R (Rstudio).
We are currently expanding IXI by building a toolchain that
supports the whole data science process, that is:
◦ data extraction
◦ feature engineering
◦ model pipeline creation
◦ experiment evaluation and comparison of different model parameters
◦ model reviewing
◦ model testing
◦ model deployment and serving
supports building and deploying models on Kubernetes as a service
supports machine learning (CPU) as well as deep learning (GPU) workloads
can be run in a hybrid setting (run on premise but scale out to cloud)
For operational excellence we build our technology stack on Kubernetes, Kubeflow, Istio, Docker, Keras + Tensorflow, GitHub and Seldon - a stack that is strong in ML and AI related operations but also close to our architecture principles and structures we live as a company.
In the end, it is absolutely vital to invest in something lasting that will push us to the next level of excellence. When establishing this new mindset throughout our company we build on quick experimentation and simple deployment.
Join us
If this sounds like an interesting environment for you and if you want to be part of our team: we are currently looking for an AI DevOps Engineer. Let’s have a chat!
*German version to be found here